PREDICTIVE POLICING: A COMPARATIVE STUDY OF THREE …

PREDICTIVE POLICING: A COMPARATIVE STUDY OF THREE HOTSPOT

MAPPING TECHNIQUES

Zachary Thomas Vavra

Submitted to the faculty of the University Graduate School

in partial fulfillment of the requirements

for the degree

Master of Science

in the Department Geography,

Indiana University

May 2015

ii

Accepted by the Graduate Faculty, Indiana University, in partial

fulfillment of the requirements for the degree of Master of Science.

Master’s Thesis Committee

_________________________________

Vijay O. Lulla Ph.D., Chair

_________________________________

Jeffrey S. Wilson, Ph.D.

_________________________________

Daniel P. Johnson, Ph.D.

iii


PREDICTIVE POLICING: A COMPARATIVE STUDY OF THREE HOTSPOT

MAPPING TECHNIQUES

Law enforcement agencies across the U.S. use maps of crime to inform their

practice and make efforts to reduce crime. Hotspot maps using historic crime data can

show practitioners concentrated areas of criminal offenses and the types of offenses that

have occurred; however, not all of these hotspot crime mapping techniques produce the

same results. This study compares three hotspot crime mapping techniques and four

crime types using the Predictive Accuracy Index (PAI) to measure the predictive

accuracy of these mapping techniques in Marion County, Indiana. Results show that the

grid hotspot mapping technique and crimes of robbery are most predictive.

Understanding the most effective crime mapping technique will allow law enforcement to

better predict and therefore prevent crimes.

Vijay O. Lulla Ph.D., Chair

iv

Table of Contents

List of Tables……………………………………………………………………………...v

List of Figures…………………………………………………………………………….vi

Introduction……………………………………………………………………………..…1

Background……………………………………………………………………………......2

Study Area……………………………………………………………………………….12

Data………………………………………………………………………………………14

Methodology……………………………………………………………………………..18

Results…………………………………………………………………………………....27

Discussion/Conclusion…………………………………………………………………...30

Appendix A……………………………………………………………………………....35

Appendix B………………………………………………………………………………36

Appendix C………………………………………………………………………………49

Appendix D………………………………………………………………………………53

Bibliography……………………………………………………………………………..55

Curriculum Vitae

v

List of Tables

Table 1. Crime Types within each Crime Group & Number of Incidents………………16

Table 2. Input Data Timeframes………………………………………………………...18

Table 3. Measurement Data Timeframes……………………………………………......19

Table 4. Grid Cell Sizes for Grid Hotspot Maps………………………………….…….24

Table 5. Mean PAI Values for Hotspot Mapping Techniques…………………………..27

Table 6. Mean PAI Values for Crime Types……………………………………………28

Table 7. Mean PAI Values for Hotspot Mapping Techniques, by Crime Type………...29

Table 8. Mean PAI Area % for each Hotspot Mapping Technique……………………..30

vi

List of Figures

Figure 1. Example of a Census Block Choropleth Hotspot Map…………………………5

Figure 2. Example of a Grid Choropleth Hotspot Map…………………………………..7

Figure 3. Visual Process of Kernel Density Estimation (KDE)………………………..…8

Figure 4. Example of a KDE Hotspot Map……………………………………………...10

Figure 5. Map of the Study Area………………………………………………………..13

1

Introduction

Hotspot mapping has become one of the most widespread methods used to

analyze and predict future crime. As of 2007, all law enforcement agencies in the U.S.

serving populations of 500,000 or more were analyzing crime using hotspot mapping

(Reaves, 2010). Hotspot maps are used to identify the areas where crimes are

concentrated, or “hot”, relative to the crime distribution in the region. Law enforcement

agencies use hotspot maps to prioritize and strategize their efforts in reducing crime. The

extensive use of hotspot mapping by law enforcement, and the multiple types of hotspot

mapping techniques available to choose from, begs the question - are all hotspot mapping

techniques equal in their ability to predict the areas where crimes may occur?

The objective of this study is twofold:

1. Compare three hotspot crime mapping techniques to see if there are differences

between the techniques’ abilities to predict where offenses may occur.

2. Determine if the accuracy of hotspot crime mapping differs between the types of

offenses being mapped.

2

Background

The origins of mapping crime can be traced back to France, where in 1829

Adriano Balbi and Anfre-Michel Guerry developed maps displaying the relationship

between the citizens’ education levels and offenses against people and property

(Weisburd & McEwen, 1998). Within 20 years crime mapping had spread to England.

In 1849 statistician Joseph Fletcher created maps comparing the rate of male

incarceration to serious property and violent crimes across counties in England and Wales

(Chamard, 2006). Although crime mapping in the U.S. did not emerge until the early 20th

century, its sophisticated use among urban sociologists further revealed the strong

connection between crime levels and the condition of the social environment.

[. . . in 1927] Frederic Thrasher superimposed the "location and

distribution" of gangs in Chicago on a map of urban areas in the city. He

found that gangs were concentrated in areas of the city where social

control was weak and social disorganization pervasive. Shaw and Myers

reached similar conclusions in [. . . their 1929] study of juvenile

delinquency conducted for the Illinois Crime Survey. . . . they show that

the home addresses of over 9,000 delinquents are clustered in areas

marked by "physical deterioration, poverty and social disorganization".

(as cited in Weisburd & McEwen, 1998, p. 8)

The first computer generated crime maps appeared in the in the mid-1960s when a

St. Louis police department mapped larcenies from automobiles. This was a great leap

forward in the advancement of crime mapping; however, the expense and expertise

required to produce these early maps limited the technology’s availability to only a

handful of law enforcement agencies. The widespread use of computerized crime

mapping did not begin until the late 1980s with the advent of the desktop computer.

Computer technology and crime mapping software have evolved significantly

over the last 30 years. Complex algorithms and high-speed processors have increased the

3

credibility of the crime fighting strategy known as predictive policing. “Predictive

policing is the application of analytical techniques – particularly quantitative techniques –

to identify lively targets for police intervention and prevent crime or solve past crimes by

making statistical predictions” (Perry, McInnis, Price, Smith, & Hollywood, 2013, pp. 1-

2). The concept of predictive policing is alluring and can conjure up fanciful ideas if not

clearly understood. “Predictions are generated through statistical calculations that

produce estimates, at best; like all techniques that extrapolate the future based on the past,

they assume that the past is prologue. Consequently, the results are probabilistic, not

certain” (Perry et al., 2013, p. 8).

While there are over a dozen predictive policing mapping techniques, this

research focused on three methods used to identify crime hotspots: jurisdiction-bounded

mapping, grid mapping, and kernel density estimation (KDE). These mapping techniques

were selected because they represent fundamental types of hotspot mapping that are

commonly discussed in crime mapping literature (S. Chainey, Tompson, & Uhlig, 2008,

p. 15). The mapping techniques examined in this study identify crime hotspots. The

term “hotspot” is widely used in crime mapping literature, but there is not a definitive

definition of what a hotspot is. A hotspot can be a specific location, such as a mall, bar,

or parking lot (Sherman, Gartin, & Buerger, 1989, p. 45); or it may adhere to strict

guidelines, such as not being more than a standard linear street block, not extending for

more than half of a block from either side of an intersection, and being at least a block

away from another hotspot (Buerger, Cohn, & Petrosino, 1995, p. 240). While ultimately

the definition of a hotspot is unique to each study, there is a common understanding that

“. . . a hotspot is an area that has a greater than average number of criminal or disorder

4

events, or an area where people have a higher than average risk of victimization” (Eck &

National Institute of Justice (U.S.), 2005, p. 2).

The jurisdiction-bounded crime mapping technique is a type of choropleth map

which aggregates the total number of offenses that occur within polygons that are created

using a certain jurisdictional boundary system (e.g., census blocks, census tracks, police

zones, etc.). Because the area of the polygons differ in size, it is necessary to normalize

the raw crime counts within each polygon by dividing them by an appropriate

denominator, such as the number of houses for burglaries, or the number of residents for

robberies (Spencer Chainey & Ratcliffe, 2005, p. 151). Each polygon is then assigned a

color based on its crime percentage rate, with darker colors typically representing

hotspots. The census block is the jurisdictional boundary system used in this study.

Figure 1 illustrates an example of a census block choropleth hotspot map.

5

Figure 1. A census block choropleth hotspot map of residential burglaries in

Indianapolis, from September 1, 2009 – February 28, 2010.

6

The grid mapping technique is another type of choropleth map. This technique

involves laying a grid of equally proportioned cells over the crime point data and

aggregating the crime points within each cell. Different colors are then assigned to each

cell based on the total number of offenses within. Unlike the jurisdiction-bounded map,

the cells are equal in area; therefore, it is not necessary to normalize the crime data.

The grid mapping technique is able to display the actual crime patterns in greater

spatial detail than the jurisdiction-bounded technique if the correct grid cell size is used;

the challenge is determining the best grid cell size. If the grid cells are too large the

resolution on the map will be coarse, making it difficult to identify the hotspots, while

grid cells that are too small produce maps which diminish the clarity of the crime patterns

and hotspots.

The literature provides various methods on how to select an appropriate grid cell

size. There is no consensus on how this process should be done because different

techniques produce more informative results depending on the data being mapped, the

application, location, etc.. Chainey and Ratcliffe (2005) suggest dividing the longest

extent of the map by 50 and using this distance as the initial grid cell size (p. 153). Hengl

(2006) professes a suitable grid resolution is determined by examining the inherent

properties of the input data and provides a series of statistical formulas to determine the

coarsest, finest, and recommended grid resolution (pp. 1295-1296). Still others examine

the physical terrain, such as the average street length, when determining the grid cell size

(Kennedy, Caplan, & Piza, 2011, p. 348). Regardless of which suggestion is

implemented, experimentation and trial and error of different grid cell sizes is often

7

required when determining the best grid cell size. Figure 2 illustrates an example of a

grid choropleth hotspot map.

Figure 2. A grid choropleth hotspot map, with 280 meter size cells, displaying

robberies in Indianapolis, from December 1, 2009 – February 28, 2010.

8

Kernel density estimation (KDE) is an interpolation mapping technique which

“smoothes” discrete crime points and creates a continuous risk surface that represents the

density or volume of crimes distributed across a study area (Eck & Justice, 2005, p. 26;

Silverman, 1986) . “The objective is to use crime incident data to identify hot spots

based on their proximity to actual crime incidents. A kernel is a standardized weighting

function used, in this application, to smooth crime incident data” (Perry et al., 2013, p.

24). The KDE process is explained in the following steps:

1. A fine grid is generated over the point distribution.

2. A moving three-dimensional function of a specified radius visits each cell

and calculates weights for each point within the kernel’s radius. Points

closer to the center will receive a higher weight, and therefore contribute

more to the cell’s total density value (Figure 3).

3. Final grid cell values are calculated by summing the values of all kernel

estimates for each location (Eck & National Institute of Justice (U.S.),

2005, pp. 26-27; Silverman, 1986).

Figure 3. The visual process of kernel density estimation. From “Kernel density

estimation and hotspot mapping,” by Hart & Zandbergen, 2014, Policing, 37(2),

p. 309.

9

KDE is regarded by many as having advantages over other crime mapping

techniques due to its growing availability in GIS software, aesthetic appearance, and

perceived accuracy with predicting crime hotspots (S. Chainey et al., 2008, p. 8). While

the KDE mapping technique is visually appealing and more statistically complex than the

other hotspot mapping techniques, it is not without its limitations. The KDE smoothing

technique tends to exaggerate the distribution of crime by spreading crime point data into

areas where crimes may not have occurred. KDE maps also require several user-defined

parameters to be set. In addition to selecting an appropriate grid cell size, the KDE

process requires users to select a suitable interpolation method and bandwidth length. An

example of a KDE hotspot map is shown in Figure 4.

10

Figure 4. A Kernel Density Estimation (KDE) hotspot map displaying crimes of

aggravated assault in Indianapolis, from March 1, 2009 – February 28, 2010.

11

All three hotspot crime mapping techniques examined in this study produce

choropleth maps which by nature suffer from the modifiable areal unit problem, or

MAUP. The MAUP is “. . . a problem arising from the imposition of artificial units of

spatial reporting on continuous geographical phenomenon resulting in the generation of

artificial spatial patterns” (Heywood, Cornelius, & Carver, 1998, p. 271). In other words,

choropleth maps often distort the cartographic appearance of the actual crime patterns

because the crimes are aggregated to polygons with arbitrary boundaries. This may result

in producing misleading hotspot maps because it “. . . shades the whole of a region and

can often be too coarse to represent the detailed spatial patterns of actual crime events”

(Spencer Chainey & Ratcliffe, 2005, p. 151).

12

Study Area

This study analyzed criminal offenses that occurred within the Indianapolis

Metropolitan Police Department’s (IMPD) six service districts, which cover an estimated

941 square kilometers. All of IMPD’s service districts are located within Marion County,

Indiana; home to Indianapolis, the 12th largest city in the U.S., with a 2010 population of

820,445 (Bureau, 2014). As a major U.S. city, the study area includes a diverse mix of

land use (residential, commercial, retail, vacant properties, etc.), demographics (ethnic

and economic), and industry. A map of the study area is displayed in Figure 5.

13

Figure 5. A map of the study area: the Indianapolis Metropolitan Police

Department’s service district in Marion County, Indiana.

14

Data

The data used in this study was from the Indianapolis Metropolitan Police

Department’s Uniform Crime Reports (UCR). The Uniform Crime Reporting Program is

administered by the FBI. It began in 1929 as a way to collect crime statistics from police

departments across the United States. The program standardizes how crime data is

submitted in order to produce uniformity in nationwide crime reporting (FBI, 2004). As

part of the standardization process, offenses were strictly defined and divided into two

groups – Part I and Part II. Part I offenses are more serious; they include eight crime

groups that are further categorized into 22 crime types. These crime groups include:

criminal homicide, forcible rape, robbery, aggravated assault, burglary, larceny-theft,

motor vehicle theft, and arson. The Part II group is made up of 21 less serious offenses.

The UCR data used in this study are criminal offenses that occurred with the

IMPD’s area of jurisdiction. Some of the offenses have been cleared by the IMPD by

arrest or exceptional means, while other offenses remain unsolved. The UCR data does

not indicate whether or not the offender was convicted of an offense. The IMPD

publishes their UCR data annually and makes it available to the public. The data was

obtained from the IMPD’s website

(http://www.indy.gov/egov/city/dps/impd/crimes/pages/ucrdownload.aspx). The data has

been geocoded using the reported addresses based on the street centerline rather than on

parcels, so the points represent an estimation of where the offenses occurred, not

necessarily an exact location. Also included in the UCR data are the date and time of

when the offense took place. If a time span was reported for a given offense, the earliest

date or time the offense could have taken place is recorded.

15

Four Part I UCR crime groups were selected for analysis: aggravated assault,

residential burglary, robbery, and vehicle theft. The study’s timeframe spans two years:

March 1, 2009 – February 28, 2011. These crime groups were selected because they are

similar to the types of offenses analyzed in a comparable 2008 study, The Utility of

Hotspot Mapping for Predicting Spatial Patterns of Crime. The authors of this study

examined residential burglary, street crime, theft from a vehicle, and theft of a vehicle,

because “. . . they are groupings that are regularly analyzed by police and crime reduction

practitioners; therefore, the implications of the research would be accessible and could be

more readily translated into policing and crime reduction practice” (S. Chainey et al.,

2008, p. 11).

Before the data could be analyzed, it had to first be cleaned and organized.

“Cleaning” the data consisted of removing unwanted offense records from the dataset.

First, records tagged with an X11 beat code were removed from the data. X11 denotes an

offense with an unknown location of occurrence. Next, simple assault offenses were

removed because they are not a Part I offense – they are a Part II offense that are included

in the Part I UCR “. . . as a quality control matter and for the purpose of looking at total

assault violence” (FBI, 2004, p. 26). Non-residential burglaries were also removed

because the data needed to normalize this type of offense was not available. Finally,

offense data that occurred outside of the study’s timeframe was removed.

The study’s two-year timeframe spans across three calendar years; therefore,

offense data was used from the IMPD’s 2009, 2010, and 2011 UCRs. Each of the three

UCRs contain offenses that occurred in a different calendar year than the UCR in which

they were reported. For example, the 2009 UCR includes a small number of offenses that

16

occurred in 2007 and 2008. Lieutenant Don Weilhamer, Jr., of the IMPD explained the

reason this happens is because occasionally the reporting process can be delayed and

miss the UCR’s monthly submission deadline. Because the IMPD is obligated to report

the offense data, the offenses are included in the next UCR submission, which is

sometimes the following calendar year (D. Weilhamer Jr., personal communication,

September 30, 2014).

The IMPD organizes their UCR data into dozens of crime categories. The four

crime groups used in this study (aggravated assault, residential burglary, robbery, and

vehicle theft) make up 28 of these crime categories. The 28 crime categories were

consolidated and organized into the four crime groups as depicted in Table 1.

Table 1

Crime Types within each Crime Group & Number of Offenses (3/1/2009 - 2/28/2011)

Crime Group: Aggravated Assault Offenses

ASSAULT – GUN 2,003

ASSAULT – HAND, FIST 2,777

ASSAULT – KNIFE 1,681

ASSAULT – OTHER WEAPON 4,584

Aggravated Assault Total 11,045

Crime Group: Residential Burglary

BURGLARY – ATTEMPT – RESIDENTIAL DAY 1,453

BURGLARY – ATTEMPT – RESIDENTIAL NIGHT 1,004

BURGLARY – FORCED ENTRY – RESIDENTIAL DAY 11,233

BURGLARY – FORCED ENTRY – RESIDENTIAL NIGHT 6,253

BURGLARY – NO FORCE – RESIDENTIAL DAY 3,046

17

BURGLARY – NO FORCE – RESIDENTIAL NIGHT 2,171

Residential Burglary Total 25,160

Crime Group: Robbery

ROBBERY – ARMED BANK 33

ROBBERY – ARMED CHAIN STORE 88

ROBBERY – ARMED COMMERICIAL HOUSE 665

ROBBERY – ARMED HIGHWAY 2,041

ROBBERY – ARMED MISCELLANEOUS 30

ROBBERY – ARMED OIL STATION 105

ROBBERY – ARMED RESIDENCE 691

ROBBERY – ATTEMPT – STRONG-ARMED 350

ROBBERY – ATTEMPT – ARMED 621

ROBBERY – STRONG-ARMED BANK 29

ROBBERY – STRONG-ARMED CHAIN STORE 23

ROBBERY – STRONG-ARMED COMMERICIAL HOUSE 345

ROBBERY – STRONG-ARMED HIGHWAY 1,438

ROBBERY – STRONG-ARMED MISCELLANEOUS 41

ROBBERY – STRONG-ARMED OIL STATION 24

ROBBERY – STRONG-ARMED RESIDENCE 624

Robbery Total 7,148

Crime Group: Vehicle Theft

VEHICLE THEFT 8,389

VEHICLE THEFT – ATTEMPT 620

Vehicle Theft Total 9,009

Total Offenses 52,362

18

Methodology

To compare the accuracy of predictive hotspot crime maps, the data was

organized chronologically and divided into “input data” and “measurement data.” The

day that separates the input data from the measurement data is referred to as the

“measurement date.” The approach to selecting a measurement date for this study

followed the methods used in Chainey et al.’s (2008, p. 11). Specifically, the

measurement date could not be a major holiday and should be representative of a day in

Indianapolis in which people go about their “normal” day-to-day routine. The

measurement date of Monday, March 1 2010 met this criteria and was selected for this

study. The input data consists of all offenses that occurred before the measurement date.

The measurement data consists of offenses that took place on and after the measurement

date. Although all of the data used in this study is historic, for this investigation the input

data was used as retrospective data while the measurement data was used as “future”

data.

The offense data was further divided into four timeframes, each three months long

(Table 2).

Table 2

Input Data Timeframes

3 Months 6 Months 9 Months 12 Months

12/1/2009 - 2/28/2010 9/1/2009 - 2/28/2010 6/1/2009 - 2/28/2010 3/1/2009 - 2/28/2010

The data in each time period is an aggregation of all offense data up to the measurement

date. In other words, the 3 months of input data include the 3 months of all offenses that

19

occurred before the measurement date; the 6 months of input data include the 6 months of

all offenses that occurred before the measurement date, and so on. The year of

measurement data was also separated into four, three month long timeframes as displayed

in Table 3. The offense data represented in each of these time periods is an aggregation of

all the offenses that occurred in the timeframe on and after the measurement date.

Table 3

Measurement Data Timeframes

3 Months 6 Months 9 Months 12 Months

3/1/2010 - 5/31/2010 3/1/2010 - 8/31/2010 3/1/2010 - 11/30/2010 3/1/2010 - 2/28/2011

There are a variety of ways to measure the predictive accuracy of hotspot

mapping techniques. Hit rate is the percentage of offenses that occur in areas where

offenses are predicted to occur (i.e. hotspots) with respect to the total number of offenses

in the dataset. The hit rate method, while useful and straightforward, fails to factor in the

size of the areas where offenses are predicted to occur – a significant shortcoming when

you consider the following: “…a hit rate could be 100 per cent, but the area where crimes

are predicted to occur could cover the entire study area – a result of little use to

practitioners who have the need to identify where to target resources” (S. Chainey et al.,

2008, p. 12).

Given this limitation, the measurement formula used in this study is the prediction

accuracy index (PAI) as first introduced by Chainey et al., 2008. According to Chainey

et al.,

This index has been devised to consider the hit rate against the areas

where crimes are predicted to occur with respect to the size of the study

20

area. The PAI is calculated by dividing the hit rate percentage (the

percentage of crime events for a measurement data time period falling into

the areas where crimes are predicted to occur determined from input data,

i.e. the crime hotspots) by the area percentage (the percentage area of the

predicted areas (the hotspots) in relation to the whole study area). (2008,

pp. 12, 14)

The formula for the PAI is as follows:

(1)

n is the number of offenses in areas where offenses are predicted to occur

(i.e. hotspots), N is the total number of offenses that occur in the study

area, a is the total areas where offenses are predicted to occur (i.e.

hotspots), and A is the total study area. (S. Chainey et al., 2008, p. 14)

Finding an equal percentage of hit rate and area percentage will produce a PAI value of

1. PAI values greater than 1 have hit rate percentages that are greater than their area

percentages, while PAI values less than 1 have hit rate percentages that are less than their

area percentages. For example, if 4% of measurement data occurs within hotspots that

make up 2% of the study area, a PAI value of 2 is produced. In short, larger PAI values

denote a greater number of future offenses occurring in hotspots that are smaller than the

study area.

In order to determine if the accuracy between hotspot maps differs between

hotspot mapping techniques, or between crime types, PAI measurements must first be

calculated. First, 48 hotspot maps were created using input data from each of the four

input data timeframes, four crime types, and three mapping techniques (4 timeframes x 4

PAItageAreaPercen

HitRate

A

a

N

n

100

100

21

crime types x 3 mapping techniques = 48 hotspot maps). This is displayed in Appendix

A.

Next, measurement data from each timeframe was laid over each of the 48 hotspot

maps. The measurement data laid over the hotspot maps allowed for the calculation of

the hit rate and area percentage, producing 192 PAI measurements. A visual of this

process is displayed in Appendix B.

To determine if there are differences in the ability of hotspot mapping techniques

to predict where crimes may occur, all of the PAI values calculated for each hotspot

mapping technique were aggregated and averaged together to produce three mean PAI

values, one for each hotspot mapping technique. This equated to averaging:

64 PAI values for the census block hotspot mapping technique

64 PAI values for the grid hotspot mapping technique

64 PAI values for the KDE hotspot mapping technique

A visual of this process is shown in Appendix C.

To find out if the ability to predict where crimes may occur differed by crime

type, all of the PAI values for each crime type were aggregated and averaged together

across the hotspot mapping techniques. This produced four mean PAI values, one for

each crime type. This equated to averaging:

48 PAI values for aggravated assault

48 PAI values for residential burglary

48 PAI values for robbery

22

48 PAI values for vehicle theft

Appendix D displays a visual of this process.

All hotspot mapping techniques require the user to set certain parameters before a

hotspot map can be produced. Given the significant influence the parameter selection has

on the resulting hotspot maps, the parameters used in this study are based on suggestions

found in literature.

ArcMap was used to create the census block choropleth hotspot maps. The 2010

U.S. census blocks, which include population and housing unit counts, were used in the

study. The point data was joined to the census blocks using the “Spatial Join” tool in

ArcMap. The default settings of the Spatial Join tool were used when creating these

maps, specifically, the “intersect match option” and the “join one-to-one join operation.”

These settings match the crime points to the census blocks with which they intersect.

Crime points located within more than one census block are not duplicated; they are

assigned to only one census block.

Because the census blocks vary in size the raw crime counts for each census block

were normalized. The normalization process consisted of dividing the number of crimes

that occurred in each census block by an appropriate denominator, which was determined

by the crime type. Residential burglaries were divided by the number of residential

housing units in each block, and crimes of aggravated assault, robbery, and vehicle theft

were divided by the number of residents in each census block (Spencer Chainey &

Ratcliffe, 2005, pp. 374-375).

23

ArcMap was also used to create the grid hotspot maps using the “fishnet” tool. A

fishnet is simply a grid made up of equal sized cells. The appropriate cell size was

determined for each input dataset based on the geometry of the point patterns,

specifically, the distance between the crime points using the following formula

introduced by Hengl (2006).

(2)

p is the grid (pixel) cell size, A is the study area in square meters, and N is the total

number of observations (i.e. crime points) (pp. 1289-1290). The 0.25 (mm) part of

the formula is one-half the distance suggested by McBrantney, Mendoca, and

Minasny who state:

. . . there should be at least 2 x 2 pixels to represent smallest rounded objects of interest and at least two pixels to represent the width of elongated objects. The smallest objects are typically of size 1 x 1 mm on the map, so that the grid resolution can be determined using the p = 0.5 mm rule. (as cited in Hengl, 2006, p. 1286)

However, the 0.5 mm rule is valid only with regular point samples, not random or

clustered distribution of points like the crime data used in this study. With random point

data “…the average spacing between closest point pairs is approximately half the spacing

between closest point pairs in regular point samples. . . . because random sampling has

equal probability of producing totally clustered and totally regular samples” (Hengl,

2006, pp. 1289-1290).

Table 4 displays the grid cell sizes used to create the grid hotspot maps based on

the number of crime points in each dataset.

p 0.25 A

N

24

Table 4

Grid cell sizes for grid hotspot maps

Assault Burglary Robbery Vehicle Theft

3 Mo Asslt Grid Map

# of Crime Points: 1088

Grid Cell Size: 233m2

3 Mo Burg Grid Map



3 Mo Robb Grid Map



3 Mo VT Grid Map

# of Crime Points:

1115

Grid Cell Size:

230m2

6 Mo Asslt Grid Map



6 Mo Burg Grid Map



6 Mo Robb Grid Map



6 Mo VT Grid Map

# of Crime Points:

2266

Grid Cell Size:

161m2

9 Mo Asslt Grid Map



9 Mo Burg Grid Map



9 Mo Robb Grid Map



9 Mo VT Grid Map

# of Crime Points:

3425

Grid Cell Size:

131m2

12 Mo Asslt Grid Map



12 Mo Burg Grid Map

# of Crime Points:

12982


12 Mo Robb Grid Map


Grid Cell Size: 124 m2

12 Mo VT Grid Map

# of Crime Points:

4476

Grid Cell Size:

115m2

The kernel density hotspot maps were created using CrimeStat III, a free spatial

statistic program for analyzing crime point locations. Unlike ArcMap, CrimeStat III

allows the user to manipulate all of the KDE parameters (method of interpolation, grid

cell size, and bandwidth length). The ability to adjust these parameters was important in

order to implement the recommendations found in literature.

Two of the three parameter settings used in this study, method of interpolation

and bandwidth length, are based on findings from Hart, and Zandbergen’s (2014)

research on the effects the interpolation method, grid cell size, and bandwidth settings

25

have on KDE hotspot maps used to forecast crime. In their study, Hart et al. adjusted

these three parameters and examined the effects the different settings had on the

predictive accuracy of the KDE hotspot maps. For the interpolation method Hart et al.

(2014) suggest using either the triangular or quartic interpolation method, depending on

the crime type, as they were methods which produced consistently high predictive

accuracy scores. This study used the triangular interpolation method for crimes of

aggravated assault, and the quartic method for robbery, residential burglary, and vehicle

theft (pp. 316-317).

In terms of selecting a bandwidth, Hart et al. (2014) recommend using a smaller

bandwidth because their ability to successfully forecast crime declined as they increased

the bandwidth search radius. The authors suggest “…a standard search radius, equal to

the smaller of the length or width of a study area, divided by between 30 and 50 be used

for determining an appropriate KDE bandwidth” (p. 316). A bandwidth of 0.9 kilometer

was used for all the KDE hotspot maps in this study. This length was determined by

dividing the shorter length of the study area (32 km) by 30 (1.07 km) and by 50

(0.64 km) and using the mean of these two calculations.

Hart et al.’s suggestion of selecting a grid cell size was not followed for this

study. Instead, the methods recommended by Hengl (2006) were implemented, and the

same grids created for the grid hotspot maps were used for the KDE hotspot maps. The

decision to use Hengl’s method over the one put forward by Hart et al. is because Hengl’s

research on the topic of selecting an appropriate grid size is more thorough.

26

The census block choropleth, grid choropleth, and kernel density hotspot mapping

techniques all produce a count of the number of crime points that fall within a defined

area. In order to determine which areas on the map are “hot,” choropleth thresholds need

to be set by selecting a classification method and the number of classes. The quantile

classification method and five classes are the choropleth thresholds used for all three of

the hotspot mapping techniques. The quantile classification method was selected for this

study because it is the choice method for comparing the different crimes types, each with

differing values (number of offenses), against each other (Santos, 2005, p. 213). It is also

the classification method often used when comparing data mapped at different time

periods, as is the data used in this study (MacEachren, 1995, p. 47). Five classes were

selected because it falls between the upper and lower limits of the suggested number of

classes to use in a choropleth map (Harries & National Institute of Justice (U.S.), 1999, p.

50).

27

Results

This study examined three hotspot crime mapping techniques commonly used to

generate hotspot maps (Weir & Bangs, 2007) to determine if differences exist between

the mapping techniques’ ability to predict the areas where offenses may occur. The study

also looked at four different crime types (aggravated assault, residential burglary,

robbery, and vehicle theft) to see if the type of crime being mapped had any influence on

the hotspot maps’ ability to predict the areas where the respective offenses may occur in

the future.

Table 5 displays the mean PAI values from the three hotspot mapping techniques.

Table 5

Mean PAI Values for Hotspot Mapping Techniques

Hotspot Mapping Technique Mean PAI Value

Jurisdiction Boundary (Census Block) 3.52

Grid 28.88

Kernel Density Estimation (KDE) 4.19

The value in bold indicates the highest PAI value and the italicized value

indicates the lowest PAI value.

The grid hotspot mapping technique proved to be the best hotspot mapping technique for

predicting the areas where offenses may occur. The results show the grid hotspot

mapping technique producing an average PAI value well above the other mapping

techniques, followed by kernel density estimation, and the census block technique having

the lowest average PAI value.

28

The research also discovered that the type of offense being mapped affects the

hotspot maps’ ability to predict the areas where the respective offenses may occur in the

future. Table 6 displays the mean PAI values, across the hotspot mapping techniques for

the four crime types examined in this study.

The value in bold denotes the highest PAI value and the italicized PAI value

indicates the lowest PAI value.

In this study the hotspot maps had a greater ability at predicting the areas where future

robberies were going to occur than the three other crime types.

To further examine how hotspot mapping techniques and crime types influence

PAI values, the mean PAI values were calculated for each hotspot mapping technique, by

crime type. Table 7 displays the consistency of these mean PAI values; with the grid

hotspot mapping technique and robbery producing the highest mean PAI values.

Table 6

Mean PAI Values for Crime Types

Crime Type Mean PAI Values

Assault 13.76

Burglary 10

Robbery 18.2

Vehicle Theft 6.82

29

Table 7

Average PAI Values for Hotspot Mapping Technique, by Crime Type

Hotspot Mapping Technique Assault Burglary Robbery Vehicle Theft

Jurisdiction Boundary (Census Block) 3.94 2.64 5.05 2.45

Grid 33.12 23.75 44.38 14.25

Kernel Density Estimation (KDE) 4.21 3.6 5.19 3.75

The values in bold indicate the highest PAI values by mapping technique, and

the italicized values indicates the highest PAI values by crime type.

30

Discussion/Conclusion

The results of this research show that not all hotspot mapping techniques are

equal in their ability to forecast the areas where crimes may occur. Not only did the grid

hotspot mapping technique consistently produce greater PAI values than the other

mapping techniques, but also its mean PAI value (28.88) is more than seven times greater

than the kernel density estimation technique’s mean PAI value (4.19). To understand

why the grid hotspot mapping technique produced PAI values that were much larger, one

must examine the area percentages within the PAI values.

As discussed earlier, PAI is the formula used in this study to measure the

predictive accuracy of the hotspot mapping techniques. All of the PAI values produced

in this study are a reflection of hit rate percentages divided by area percentages. When

examining these factors it becomes clear that the reason the grid hotspot mapping

technique produced PAI values that are so much greater is because of how small the

mean area percentage is for the grid hotspot maps compared to the other hotspot mapping

techniques (Table 8).

Table 8

Mean PAI Area % for each Hotspot Mapping Technique

Hotspot Mapping Technique Mean Area %

Census Block 2.56%

Grid 0.11%

KDE 16.31%

The reason the mean area percentages differ between the hotspot mapping

techniques is because each technique identifies hotspots differently. When the census

block hotspot map identifies a “hot” census block, the area is likely much larger than

31

when the grid hotspot map identifies a “hot” cell due to the great difference in the

average size of these areas. The mean area of the census blocks in this study is 62,364

square meters, compared to 116.5 square meters for the grid hotspot mapping technique.

KDE identifies the “hot” areas on a map by looking at the high density of crime in

relation to the rest of the study area. Although the same grids used in the grid hotspot

maps were also used to make the KDE maps, the KDE hotspots are much larger than the

hotspots produced by the other mapping techniques due to KDE’s “smoothing” process

which tends to exaggerate the “hot” areas.

Another factor that has influenced the results of this study is the size of the study

area. This becomes clear when comparing this study’s results to Chainey et al.’s, which

resulted in KDE having the highest mean PAI value. The crime types, number of crimes

analyzed, hotspot mapping techniques used, and the timeframes of the two studies are

similar. What is significantly different is the size of the study area in each study. This

study’s study area (941 square kilometers) is about 25 times larger than Chainey et al.’s

study area (37 square kilometers). Given that the study area is the denominator of the

area percentage in the PAI formula, it has a significant influence on the PAI value and is

likely a contributing factor as to why the two studies show differing results in which

hotspot crime mapping techniques yielded the highest mean PAI value.

The findings also revealed that the accuracy of hotspot mapping used to predict

the areas where offenses may occur in the future differs between the types of offenses

being mapped. One of the limitations of this study is that it did not examine the

influences socio-economic levels or environmental characteristics may have had on the

hotspots that were mapped, so it is not possible to know for certain how these factors

32

affected the mean PAI values for each type of crime mapped; however, it is possible to

speculate factors which may have contributed to the differences. In Chainey et al.’s

study, street crime (robbery of personal property and theft from a person) yielded the

highest mean PAI value and vehicle theft the lowest. According to Chainey et al.:

. . . street crime predominantly occurred in areas where shops, bars,

restaurants, markets and other forms of retail and entertainment

concentrate – places that are prone to the opportunities to commit street

crime. This type of land use tends to be clustered at particular localities,

meaning that the opportunity for street crime is similarly highly

concentrated. These types of land use also tend to be static, in that they do

not shift around the urban landscape but instead become a stationary part

of the area’s environmental fabric. (2008, p. 24)

These components, coupled with the understanding that crime patterns tend to

highlight the areas which allow for opportunities to commit crime, are likely the main

reasons the hotspot maps of street crime resulted in the highest mean PAI value in

Chainey et al.’s study. One could speculate this same rational can explain why the

robberies in this study produced the highest PAI values given the similar nature of the

crime types. Vehicle thefts, as opposed to robberies, are transient and more dispersed

across the study area; therefore, these types of crime patterns are continually shifting

making the retrospective data less reliable in terms of predicting where future offenses

may occur.

Other limitations to this study involve the availability of certain data and the

method by which the data was organized. This study was unable to examine non-

residential burglaries because the non-residential unit data, necessary to normalize the

census block burglary hotspot maps, was not readily available. Secondly, the

organization of the tables in this study does not follow the “tidy data” rules as described

33

by Hadley Wickham, where each variable is a column, each observation is a row, and

each type of observational unit forms a table (2014, p. 4). The benefit of having the data

organized in this way is that “. . . it makes it easy for an analyst or computer to extract

needed variables because it provides a standard way of structuring a dataset” (Wickham,

2014, p. 5). Not having the data organized the “tidy” way slows the time it takes to

analyze the data and increases to potential for errors.

This study has shown the most predictive hotspot crime mapping technique and

predictive type of crime using hotspot mapping in Marion County, Indiana. Hotspot

crime mapping is just one technique used by law enforcement to strategize their crime-

fighting efforts. Advances in technology have built on the concepts used in hotspot crime

mapping and contributed to the growth of the predictive policing industry. An example

of this is PredPol, a leading predictive policing company with proven results. Similar to

hotspot mapping it relies on historic data (type of crime, place of crime, and time of

crime); however, rather than simply mapping the data, the data is also analyzed using a

unique algorithm based on criminal behavior patterns to highlight the areas the police

officers should patrol at any given time of the day.

PredPol reduces the amount of time crime analysts are spending looking over the

data and creating maps, allowing more time for law enforcement to put the predictive

policing techniques into practice. Multiple cities using PredPol have shown a reduction

in crime, including the Los Angeles Police Department’s Foothill Division which “. . .

saw a 20% drop in predicted crimes year over year from January 2013 to January 2014”

(PredPol, 2014). Improved predictive mapping tools, such as PredPol, are needed by

34

practitioners to produce quicker and more standardized ways of analyzing crime data to

effectively reduce crime.

35

Appendix A

48 Hotspot Crime Maps

Mapping Technique: Census Block (CB)


3 Mo (12/1/2009 -

2/28/2010)

3 Mo Assault CB

Map

3 Mo Burglary CB

Map

3 Mo Robbery CB

Map

3 Mo VT CB

Map

6 Mo (9/1/2009 -

2/28/2010)

6 Mo Assault CB

Map

6 Mo Burglary CB

Map

6 Mo Robbery CB

Map

6 Mo VT CB

Map

9 Mo (6/1/2009 -

2/28/2010)

9 Mo Assault CB

Map

9 Mo Burglary CB

Map

9 Mo Robbery CB

Map

9 Mo VT CB

Map

12 Mo (3/1/2009 -

2/28/2010)

12 Mo Assault

CB Map

12 Mo Burglary

CB Map

12 Mo Robbery

CB Map

12 Mo VT CB

Map

Hotspot Mapping Technique: Grid


3 Mo (12/1/2009 -

2/28/2010)

3 Mo Assault

Grid Map

3 Mo Burglary

Grid Map

3 Mo Robbery

Grid Map

3 Mo VT Grid

Map

6 Mo (9/1/2009 -

2/28/2010)

6 Mo Assault

Grid Map

6 Mo Burglary

Grid Map

6 Mo Robbery

Grid Map

6 Mo VT Grid

Map

9 Mo (6/1/2009 -

2/28/2010)

9 Mo Assault

Grid Map

9 Mo Burglary

Grid Map

9 Mo Robbery

Grid Map

9 Mo VT Grid

Map

12 Mo (3/1/2009 -

2/28/2010)

12 Mo Assault

Grid Map

12 Mo Burglary

Grid Map

12 Mo Robbery

Grid Map

12 Mo VT

Grid Map

Hotspot Mapping Technique: Kernel Density Estimation (KDE)


3 Mo (12/1/2009 -

2/28/2010)

3 Mo Assault KD

Map

3 Mo Burglary KD

Map

3 Mo Robbery KD

Map

3 Mo VT KD

Map

6 Mo (9/1/2009 -

2/28/2010)

6 Mo Assault KD

Map

6 Mo Burglary KD

Map

6 Mo Robbery KD

Map

6 Mo VT KD

Map

9 Mo (6/1/2009 -

2/28/2010)

9 Mo Assault KD

Map

9 Mo Burglary KD

Map

9 Mo Robbery KD

Map

9 Mo VT KD

Map

12 Mo (3/1/2009 -

2/28/2010)

12 Mo Assault

KD Map

12 Mo Burglary

KD Map

12 Mo Robbery

KD Map

12 Mo VT KD

Map

36

Appendix B

192 PAI Calculations: Measurement Data over Hotspot Crime Maps

Hotspot Mapping Technique: Census Block (CB)


Input Crime Data (ICD)

3 Mo (12/1/2009 - 2/28/2010)

3 Mo Asslt CB

Map 3 Mo Burg CB Map

3 Mo Robb CB

Map 3 Mo VT CB Map

6 Mo (9/1/2009 - 2/28/2010)

6 Mo Asslt CB


6 Mo Robb CB

Map 6 Mo VT CB Map

9 Mo (6/1/2009 - 2/28/2010)

9 Mo Asslt CB


9 Mo Robb CB

Map 9 Mo VT CB Map

12 Mo (3/1/2009 - 2/28/2010)

12 Mo Asslt CB

Map

12 Mo Burg CB

Map

12 Mo Robb CB

Map 12 Mo VT CB Map

Measurement Crime Data

(MCD)

3 Mo MCD (3/1/2010 -

5/31/2010)

3 Mo Assault

MCD

3 Mo Burglary

MCD

3 Mo Robbery

MCD

3 Mo Vehicle Theft

MCD

3 Mo Asslt CB

Map

n = 31

N = 1528 Hit rate % = .02

a = 8.99 A = 940.77

Area % = .01

PAI = 2

3 Mo Burg CB Map

n = 124

N = 2857

Hit rate % = .04

a = 15.02

A = 940.77 Area % = .02

PAI = 2

3 Mo Robb CB

Map

n = 31

N = 810 Hit rate % = .04

a = 8.16 A = 940.77

Area % = .01

PAI = 4

3 Mo VT CB Map

n = 20

N = 1001 Hit rate % = .02

a = 12.58 A = 940.77

Area % = .01

PAI = 2

6 Mo Asslt CB

Map

n = 99

N = 1528

Hit rate % = .06

a = 14.99 A = 940.77

Area % = .02

PAI = 3

6 Mo Burg CB Map

n = 270 N = 2857

Hit rate % = .09

a = 30.97

A = 940.77 Area % = .03

PAI = 3

6 Mo Robb CB

Map

n = 74

N = 810

Hit rate % = .09

a = 15.13 A = 940.77

Area % = .02

PAI = 4.5

6 Mo VT CB Map

n = 45

N = 1001

Hit rate % = .04

a = 21.44 A = 940.77

Area % = .02

PAI = 2

9 Mo Asslt CB

Map

n = 161

N = 1528

Hit rate % = .11

a = 19.35

A = 940.77 Area % = .02

PAI = 5.5

9 Mo Burg CB Map

n = 350

N = 2857

Hit rate % = .12

a = 42.62

A = 940.77

Area % = .05

PAI = 2.4

9 Mo Robb CB

Map

n = 99

N = 810

Hit rate % = .12

a = 19.72

A = 940.77 Area % = .02

PAI = 6

9 Mo VT CB Map

n = 71

N = 1001

Hit rate % = .07

a = 30.69

A = 940.77 Area % = .03

PAI = 2.33

37

12 Mo Asslt CB

Map

n = 208

N = 1528 Hit rate % = .14

a = 27.14 A = 940.77

Area % = .03

PAI = 4.66

12 Mo Burg CB

Map

n = 422

N = 2857 Hit rate % = .15

a = 46.15 A = 940.77

Area % = .05

PAI = 3

12 Mo Robb CB

Map

n = 129

N = 810 Hit rate % = .16

a = 25.26 A = 940.77

Area % = .03

PAI = 5.33

12 Mo VT CB Map

n = 90

N = 1001 Hit rate % = .09

a = 34.59 A = 940.77

Area % = .04

PAI = 2.25

6 Mo MCD (3/1/2010 -

8/31/2010)

6 Mo Assault

MCD

6 Mo Burglary

MCD

6 Mo Robbery

MCD

6 Mo Vehicle Theft

MCD

3 Mo Asslt CB Map

n = 72 N = 3155

Hit rate % = .02

a = 8.99

A = 940.77

Area % = .01

PAI = 2

3 Mo Burg CB Map

n = 253

N = 6089 Hit rate % = .04

a = 15.02 A = 940.77

Area % = .02

PAI = 2

3 Mo Robb CB Map

n = 68 N = 1663

Hit rate % = .04

a = 8.16

A = 940.77

Area % = .01

PAI = 4

3 Mo VT CB Map

n = 62 N = 2200

Hit rate % = .03

a = 12.58

A = 940.77

Area % = .01

PAI = 3

6 Mo Asslt CB

Map

n = 209

N = 3155 Hit rate % = .07

a = 14.99 A = 940.77

Area % = .02

PAI = 3.5

6 Mo Burg CB Map

n = 541

N = 6089

Hit rate % = .09

a = 30.97

A = 940.77 Area % = .03

PAI = 3

6 Mo Robb CB

Map

n = 157

N = 1663 Hit rate % = .09

a = 15.13 A = 940.77

Area % = .02

PAI = 4.5

6 Mo VT CB Map

n = 100

N = 2200 Hit rate % = .05

a = 21.44 A = 940.77

Area % = .02

PAI = 2.5

9 Mo Asslt CB

Map

n = 326

N = 3155 Hit rate % = .10

a = 19.35 A = 940.77

Area % = .02

PAI = 5

9 Mo Burg CB Map

n = 767

N = 6089

Hit rate % = .13

a = 42.62

A = 940.77 Area % = .05

PAI = 2.6

9 Mo Robb CB

Map

n = 217

N = 1663 Hit rate % = .13

a = 19.72 A = 940.77

Area % = .02

PAI = 6.5

9 Mo VT CB Map

n = 164

N = 2200 Hit rate % = .07

a = 30.69 A = 940.77

Area % = .03

PAI = 2.33

38

12 Mo Asslt CB

Map

n = 424

N = 3155 Hit rate % = .13

a = 27.14 A = 940.77

Area % = .03

PAI = 4.33

12 Mo Burg CB

Map

n = 902

N = 6089 Hit rate % = .15

a = 46.15 A = 940.77

Area % = .05

PAI = 3

12 Mo Robb CB

Map

n = 270

N = 1663 Hit rate % = .16

a = 25.26 A = 940.77

Area % = .03

PAI = 5.33

12 Mo VT CB Map

n = 207

N = 2200 Hit rate % = .09

a = 34.59 A = 940.77

Area % = .04

PAI = 2.25

9 Mo MCD (3/1/2010 -

11/30/2010)

9 Mo Assault

MCD

9 Mo Burglary

MCD

9 Mo Robbery

MCD

9 Mo Vehicle Theft

MCD

3 Mo Asslt CB Map

n = 125 N = 4638

Hit rate % = .03

a = 8.99

A = 940.77

Area % = .01

PAI = 3

3 Mo Burg CB Map

n = 390

N = 9597 Hit rate % = .04

a = 15.02 A = 940.77

Area % = .02

PAI = 2

3 Mo Robb CB Map

n = 105 N = 2609

Hit rate % = .04

a = 8.16

A = 940.77

Area % = .01

PAI = 4

3 Mo VT CB Map

n = 96 N = 3418

Hit rate % = .03

a = 12.58

A = 940.77

Area % = .01

PAI = 3

6 Mo Asslt CB

Map

n = 311

N = 4638 Hit rate % = .07

a = 14.99 A = 940.77

Area % = .02

PAI = 3.5

6 Mo Burg CB Map

n = 873

N = 9597

Hit rate % = .09

a = 30.97

A = 940.77 Area % = .03

PAI = 3

6 Mo Robb CB

Map

n = 243

N = 2609 Hit rate % = .09

a = 15.13 A = 940.77

Area % = .02

PAI = 4.5

6 Mo VT CB Map

n = 167

N = 3418 Hit rate % = .05

a = 21.44 A = 940.77

Area % = .02

PAI = 2.5

9 Mo Asslt CB

Map

n = 492

N = 4638 Hit rate % = .11

a = 19.35 A = 940.77

Area % = .02

PAI = 5.5

9 Mo Burg CB Map

n = 1207

N = 9597

Hit rate % = .13

a = 42.62

A = 940.77 Area % = .05

PAI = 2.6

9 Mo Robb CB

Map

n = 329

N = 2609 Hit rate % = .13

a = 19.72 A = 940.77

Area % = .02

PAI = 6.5

9 Mo VT CB Map

n = 263

N = 3418 Hit rate % = .08

a = 30.69 A = 940.77

Area % = .03

PAI = 2.67

39

12 Mo Asslt CB

Map

n = 633

N = 4638 Hit rate % = .14

a = 27.14 A = 940.77

Area % = .03

PAI = 4.66

12 Mo Burg CB

Map

n = 1415

N = 9597 Hit rate % = .15

a = 46.15 A = 940.77

Area % = .05

PAI = 3

12 Mo Robb CB

Map

n = 416

N = 2609 Hit rate % = .16

a = 25.26 A = 940.77

Area % = .03

PAI = 5.33

12 Mo VT CB Map

n = 318

N = 3418 Hit rate % = .09

a = 34.59 A = 940.77

Area % = .04

PAI = 2.25

12 Mo MCD (3/1/2010 -

2/28/2011)

12 Mo Assault

MCD

12 Mo Burglary

MCD

12 Mo Robbery

MCD

12 Mo Vehicle Theft

MCD

3 Mo Asslt CB Map

n = 159 N = 5699

Hit rate % = .03

a = 8.99

A = 940.77

Area % = .01

PAI = 3

3 Mo Burg CB Map

n = 500

N = 12178 Hit rate % = .04

a = 15.02 A = 940.77

Area % = .02

PAI = 2

3 Mo Robb CB Map

n = 141 N = 3349

Hit rate % = .04

a = 8.16

A = 940.77

Area % = .01

PAI = 4

3 Mo VT CB Map

n = 134

N = 4533 Hit rate % = .03

a = 12.58 A = 940.77

Area % = .01

PAI = 3

6 Mo Asslt CB

Map

n = 377

N = 5699

Hit rate % = .07

a = 14.99

A = 940.77 Area % = .02

PAI = 3.5

6 Mo Burg CB Map

n = 1121 N = 12178

Hit rate % = .09

a = 30.97

A = 940.77

Area % = .03

PAI = 3

6 Mo Robb CB

Map

n = 304

N = 3349

Hit rate % = .09

a = 15.13

A = 940.77 Area % = .02

PAI = 4.5

6 Mo VT CB Map

n = 221 N = 4533

Hit rate % = .05

a = 21.44

A = 940.77

Area % = .02

PAI = 2.5

9 Mo Asslt CB Map

n = 611 N = 5699

Hit rate % = .11

a = 19.35

A = 940.77

Area % = .02

PAI = 5.5

9 Mo Burg CB Map

n = 1543

N = 12178 Hit rate % = .13

a = 42.62 A = 940.77

Area % = .05

PAI = 2.6

9 Mo Robb CB Map

n = 428 N = 3349

Hit rate % = .13

a = 19.72

A = 940.77

Area % = .02

PAI = 6.5

9 Mo VT CB Map

n = 331

N = 4533 Hit rate % = .07

a = 30.69 A = 940.77

Area % = .03

PAI = 2.33

40

12 Mo Asslt CB

Map

n = 768

N = 5699 Hit rate % = .13

a = 27.14 A = 940.77

Area % = .03

PAI = 4.33

12 Mo Burg CB

Map

n = 1799

N = 12178 Hit rate % = .15

a = 46.15 A = 940.77

Area % = .05

PAI = 3

12 Mo Robb CB

Map

n = 549

N = 3349 Hit rate % = .16

a = 25.26 A = 940.77

Area % = .03

PAI = 5.33

12 Mo VT CB Map

n = 404

N = 4533

Hit rate % = .09

a = 34.59

A = 940.77 Area % = .04

PAI = 2.25

Hotspot Mapping Technique:

Grid



3 Mo (12/1/2009 - 2/28/2010)

3 Mo Asslt Grid

Map

# of Points: 1088

Grid Cell Size: 233m

3 Mo Burg Grid

Map

# of Points: 2554


3 Mo Robb Grid

Map

# of Points: 748


3 Mo VT Grid Map

# of Points: 1115


6 Mo (9/1/2009 - 2/28/2010)

6 Mo Asslt Grid

Map

# of Points: 2437

Grid Cell Size:

155m

6 Mo Burg Grid

Map

# of Points: 6333

Grid Cell Size: 96m

6 Mo Robb Grid

Map

# of Points: 1763

Grid Cell Size:

183m

6 Mo VT Grid Map

# of Points: 2266 Grid Cell Size: 161m

9 Mo (6/1/2009 - 2/28/2010)

9 Mo Asslt Grid

Map

# of Points: 3940

Grid Cell Size:

122 m

9 Mo Burg Grid

Map

# of Points: 9943

Grid Cell Size: 77m

9 Mo Robb Grid

Map

# of Points: 2884

Grid Cell Size:

143m

9 Mo VT Grid Map


12 Mo (3/1/2009 - 2/28/2010)

12 Mo Asslt Grid Map

# of Points: 5346 Grid Cell Size:

105 m

12 Mo Burg Grid Map


12 Mo Robb Grid Map

# of Points: 3799 Grid Cell Size:

124m

12 Mo VT Grid Map

# of Points: 4476

Grid Cell Size: 115


(MCD)

3 Mo MCD (3/1/2010 -

5/31/2010)

3 Mo Assault

MCD

3 Mo Burglary

MCD

3 Mo Robbery

MCD

3 Mo Vehicle Theft

MCD

3 Mo Asslt Grid

Map

n = 39

N = 1528

Hit rate % = .03

a = 1.19 A = 941

Area % = .001

PAI = 30

3 Mo Burg Grid

Map

n = 39

N = 2857

Hit rate % = .01

a = .83

A = 941

Area % = .001

PAI = 10

3 Mo Robb Grid

Map

n = 12

N = 810

Hit rate % = .01

a = .55 A = 941

Area % = .001

PAI = 10

3 Mo VT Grid Map

n = 6

N = 1001

Hit rate % = .006

a = .58 A = 941

Area % = .001

PAI = 6

41

6 Mo Asslt Grid

Map

n = 61

N = 1528 Hit rate % = .04

a = 1.25 A = 941

Area % = .001

PAI = 40

6 Mo Burg Grid

Map

n = 69 N = 2857

Hit rate % = .02

a = .83

A = 941

Area % = .001

PAI = 20

6 Mo Robb Grid

Map

n = 28

N = 810 Hit rate % = .03

a = 1.14 A = 941

Area % = .001

PAI = 30

6 Mo VT Grid Map

n = 10

N = 1001 Hit rate % = .01

a = .83 A = 941

Area % = .001

PAI = 10

9 Mo Asslt Grid

Map

n = 77

N = 1528

Hit rate % = .05

a = 1.55 A = 941

Area % = .002

PAI = 25

9 Mo Burg Grid

Map

n = 115

N = 2857

Hit rate % = .04

a = 1.16

A = 941

Area % = .001

PAI = 40

9 Mo Robb Grid

Map

n = 46

N = 810

Hit rate % = .06

a = .90 A = 941

Area % = .001

PAI = 60

9 Mo VT Grid Map

n = 16

N = 1001

Hit rate % = .02

a = .72 A = 941

Area % = .001

PAI = 20

12 Mo Asslt Grid

Map

n = 129

N = 1528

Hit rate % = .08

a = 1.97

A = 941 Area % = .002

PAI = 40

12 Mo Burg Grid

Map

n = 152

N = 2857

Hit rate % = .05

a = 1.45

A = 941 Area % = .002

PAI = 25

12 Mo Robb Grid

Map

n = 65

N = 810

Hit rate % = .08

a = 1.28

A = 941 Area % = .001

PAI = 80

12 Mo VT Grid Map

n = 28

N = 1001

Hit rate % = .03

a = .85

A = 941 Area % = .001

PAI = 30

6 Mo MCD (3/1/2010 -

8/31/2010)

6 Mo Assault

MCD

6 Mo Burglary

MCD

6 Mo Robbery

MCD

6 Mo Vehicle Theft

MCD

3 Mo Asslt Grid

Map

n = 82

N = 3155 Hit rate % = .03

a = 1.19 A = 941

Area % = .001

PAI = 30

3 Mo Burg Grid

Map

n = 90

N = 6089 Hit rate % = .01

a = .83 A = 941

Area % = .001

PAI = 10

3 Mo Robb Grid

Map

n = 24

N = 1663 Hit rate % = .01

a = .55 A = 941

Area % = .001

PAI = 10

3 Mo VT Grid Map

n = 8

N = 2200 Hit rate % = .004

a = .58 A = 941

Area % = .001

PAI = 4

42

6 Mo Asslt Grid

Map

n = 105

N = 3155 Hit rate % = .03

a = 1.25 A = 941

Area % = .001

PAI = 30

6 Mo Burg Grid

Map

n = 142

N = 6089 Hit rate % = .02

a = .83 A = 941

Area % = .001

PAI = 20

6 Mo Robb Grid

Map

n = 58

N = 1663 Hit rate % = .03

a = 1.14 A = 941

Area % = .001

PAI = 30

6 Mo VT Grid Map

n = 27

N = 2200 Hit rate % = .01

a = .83 A = 941

Area % = .001

PAI = 10

9 Mo Asslt Grid

Map

n = 154

N = 3155

Hit rate % = .05

a = 1.55 A = 941

Area % = .002

PAI = 25

9 Mo Burg Grid

Map

n = 252

N = 6089

Hit rate % = .04

a = 1.16 A = 941

Area % = .001

PAI = 40

9 Mo Robb Grid

Map

n = 85

N = 1663

Hit rate % = .05

a = .90 A = 941

Area % = .001

PAI = 50

9 Mo VT Grid Map

n = 35

N = 2200

Hit rate % = .02

a = .72 A = 941

Area % = .001

PAI = 20

12 Mo Asslt Grid

Map

n = 237

N = 3155

Hit rate % = .08

a = 1.97

A = 941 Area % = .002

PAI = 40

12 Mo Burg Grid

Map

n = 309

N = 6089

Hit rate % = .05

a = 1.45

A = 941 Area % = .002

PAI = 25

12 Mo Robb Grid

Map

n = 128

N =1663

Hit rate % = .08

a = 1.28

A = 941 Area % = .001

PAI = 80

12 Mo VT Grid Map

n = 53

N = 2200

Hit rate % = .02

a = .85

A = 941 Area % = .001

PAI = 20

9 Mo MCD (3/1/2010 -

11/30/2010)

9 Mo Assault

MCD

9 Mo Burglary

MCD

9 Mo Robbery

MCD

9 Mo Vehicle Theft

MCD

3 Mo Asslt Grid

Map

n = 126

N = 4638 Hit rate % = .03

a = 1.19 A = 941

Area % = .001

PAI = 30

3 Mo Burg Grid

Map

n = 137

N = 9597 Hit rate % = .01

a = .83 A = 941

Area % = .001

PAI = 10

3 Mo Robb Grid

Map

n = 31

N = 2609 Hit rate % = .01

a = .55 A = 941

Area % = .001

PAI = 10

3 Mo VT Grid Map

n = 14

N = 3418 Hit rate % = .004

a = .58 A = 941

Area % = .001

PAI = 4

43

6 Mo Asslt Grid

Map

n = 173

N = 4638 Hit rate % = .04

a = 1.25 A = 941

Area % = .001

PAI = 40

6 Mo Burg Grid

Map

n = 200

N = 9597 Hit rate % = .02

a = .83 A = 941

Area % = .001

PAI = 20

6 Mo Robb Grid

Map

n = 95

N = 2609 Hit rate % = .04

a = 1.14 A = 941

Area % = .001

PAI = 40

6 Mo VT Grid Map

n = 40

N = 3418 Hit rate % = .01

a = .83 A = 941

Area % = .001

PAI = 10

9 Mo Asslt Grid

Map

n = 229

N = 4638

Hit rate % = .05

a = 1.55 A = 941

Area % = .002

PAI = 25

9 Mo Burg Grid

Map

n = 346

N = 9597

Hit rate % = .04

a = 1.16 A = 941

Area % = .001

PAI = 40

9 Mo Robb Grid

Map

n = 141

N = 2609

Hit rate % = .05

a = .90 A = 941

Area % = .001

PAI = 50

9 Mo VT Grid Map

n = 64

N = 3418

Hit rate % = .02

a = .72 A = 941

Area % = .001

PAI = 20

12 Mo Asslt Grid

Map

n = 357

N = 4638

Hit rate % = .08

a = 1.97

A = 941 Area % = .002

PAI = 40

12 Mo Burg Grid

Map

n = 451

N = 9597

Hit rate % = .05

a = 1.45

A = 941 Area % = .002

PAI = 25

12 Mo Robb Grid

Map

n = 206

N = 2609

Hit rate % = .08

a = 1.28

A = 941 Area % = .001

PAI = 80

12 Mo VT Grid Map

n = 85

N = 3418

Hit rate % = .02

a = .85

A = 941 Area % = .001

PAI = 20

12 Mo MCD (3/1/2010 -

2/28/2011)

12 Mo Assault

MCD

12 Mo Burglary

MCD

12 Mo Robbery

MCD

12 Mo Vehicle Theft

MCD

3 Mo Asslt Grid

Map

n = 163

N = 5699 Hit rate % = .03

a = 1.19 A = 941

Area % = .001

PAI = 30

3 Mo Burg Grid

Map

n = 172

N = 12178 Hit rate % = .01

a = .83 A = 941

Area % = .001

PAI = 10

3 Mo Robb Grid

Map

n = 42

N = 3349 Hit rate % = .01

a = .55 A = 941

Area % = .001

PAI = 10

3 Mo VT Grid Map

n = 19

N = 4533 Hit rate % = .004

a = .58 A = 941

Area % = .001

PAI = 4

44

6 Mo Asslt Grid

Map

n = 232

N = 5699 Hit rate % = .04

a = 1.25 A = 941

Area % = .001

PAI = 40

6 Mo Burg Grid

Map

n = 259

N = 12178 Hit rate % = .02

a = .83 A = 941

Area % = .001

PAI = 20

6 Mo Robb Grid

Map

n = 122

N = 3349 Hit rate % = .04

a = 1.14 A = 941

Area % = .001

PAI = 40

6 Mo VT Grid Map

n = 55

N = 4533 Hit rate % = .01

a = .83 A = 941

Area % = .001

PAI = 10

9 Mo Asslt Grid

Map

n = 289

N = 5699

Hit rate % = .05

a = 1.55 A = 941

Area % = .002

PAI = 25

9 Mo Burg Grid

Map

n = 432

N = 12178

Hit rate % = .04

a = 1.16 A = 941

Area % = .001

PAI = 40

9 Mo Robb Grid

Map

n = 178

N = 3349

Hit rate % = .05

a = .90 A = 941

Area % = .001

PAI = 50

9 Mo VT Grid Map

n = 80

N = 4533

Hit rate % = .02

a = .72 A = 941

Area % = .001

PAI = 20

12 Mo Asslt Grid

Map

n = 454

N = 5699

Hit rate % = .08

a = 1.97

A = 941 Area % = .002

PAI = 40

12 Mo Burg Grid

Map

n = 567

N = 12178

Hit rate % = .05

a = 1.45

A = 941 Area % = .002

PAI = 25

12 Mo Robb Grid

Map

n = 268

N = 3349

Hit rate % = .08

a = 1.28

A = 941 Area % = .001

PAI = 80

12 Mo VT Grid Map

n = 105

N = 4533

Hit rate % = .02

a = .85

A = 941 Area % = .001

PAI = 20

Hotspot Mapping Technique: Kernel Density (KD)



3 Mo (12/1/2009 - 2/28/2010)

3 Mo Asslt KD

Map 3 Mo Burg KD Map

3 Mo Robb KD

Map 3 Mo VT KD Map

6 Mo (9/1/2009 - 2/28/2010) 6 Mo Asslt KD

Map 6 Mo Burg KD Map 6 Mo Robb KD

Map 6 Mo VT KD Map

9 Mo (6/1/2009 - 2/28/2010)

9 Mo Asslt KD

Map 9 Mo Burg KD Map

9 Mo Robb KD

Map 9 Mo VT KD Map

12 Mo (3/1/2009 - 2/28/2010) 12 Mo Asslt KD

Map 12 Mo Burg KD

Map 12 Mo Robb KD

Map 12 Mo VT KD Map


(MCD)

3 Mo MCD (3/1/2010 -

5/31/2010)

3 Mo Assault

MCD

3 Mo Burglary

MCD

3 Mo Robbery

MCD

3 Mo Vehicle Theft

MCD

45

3 Mo Asslt KD

Map

n = 888

N = 1528 Hit rate % = .58

a = 118.4 A = 941

Area % = .13

PAI = 4.46

3 Mo Burg KD Map

n = 1836

N = 2857

Hit rate % = .64

a = 162.43

A = 941 Area % = .17

PAI = 3.76

3 Mo Robb KD

Map

n = 457

N = 810 Hit rate % = .56

a = 92.61 A = 941

Area % = .1

PAI = 5.6

3 Mo VT KD Map

n = 524 N = 1001

Hit rate % = .52

a = 124.33

A = 941

Area % = .13

PAI = 4

6 Mo Asslt KD

Map

n = 1077

N = 1528

Hit rate % = .7

a = 150.77 A = 941

Area % = .16

PAI = 4.38

6 Mo Burg KD Map

n = 2023

N = 2857

Hit rate % = .71

a = 189.68

A = 941 Area % = .20

PAI = 3.55

6 Mo Robb KD

Map

n = 552

N = 810

Hit rate % = .68

a = 120.96 A = 941

Area % = .13

PAI = 5.23

6 Mo VT KD Map

n = 619

N = 1001

Hit rate % = .62

a = 155.25 A = 941

Area % = .16

PAI = 3.88

9 Mo Asslt KD

Map

n = 1144

N = 1528

Hit rate % = .75

a = 167.19

A = 941 Area % = .18

PAI = 4.17

9 Mo Burg KD Map

n = 2043 N = 2857

Hit rate % = .72

a = 189.51

A = 941

Area % = .20

PAI = 3.6

9 Mo Robb KD

Map

n = 590

N = 810

Hit rate % = .73

a = 136.02

A = 941 Area % = .14

PAI = 5.21

9 Mo VT KD Map

n = 667

N = 1001

Hit rate % = .67

a = 170.85

A = 941 Area % = .18

PAI = 3.72

12 Mo Asslt KD Map

n = 1167

N = 1528

Hit rate % = .76

a = 174.31

A = 941 Area % = .19

PAI = 4

12 Mo Burg KD Map

n = 2047

N = 2857

Hit rate % = .72

a = 189.4

A = 941 Area % = .20

PAI = 3.6

12 Mo Robb KD Map

n = 600

N = 810

Hit rate % = .74

a = 142.89

A = 941 Area % = .15

PAI = 4.93

12 Mo VT KD Map

n = 692

N = 1001

Hit rate % = .69

a = 179.94

A = 941 Area % = .19

PAI = 3.63

6 Mo MCD (3/1/2010 -

8/31/2010)

6 Mo Assault

MCD

6 Mo Burglary

MCD

6 Mo Robbery

MCD

6 Mo Vehicle Theft

MCD

46

3 Mo Asslt KD

Map

n = 1814

N = 3155 Hit rate % = .57

a = 118.4 A = 941

Area % = .13

PAI = 4.38

3 Mo Burg KD Map

n = 3923

N = 6089

Hit rate % = .64

a = 162.43

A = 941 Area % = .17

PAI = 3.76

3 Mo Robb KD

Map

n = 913

N = 1663 Hit rate % = .55

a = 92.61 A = 941

Area % = .1

PAI = 5.5

3 Mo VT KD Map

n = 1107

N = 2200 Hit rate % = .5

a = 124.33 A = 941

Area % = .13

PAI = 3.85

6 Mo Asslt KD

Map

n = 2202

N = 3155

Hit rate % = .7

a = 150.77 A = 941

Area % = .16

PAI = 4.38

6 Mo Burg KD Map

n = 4325

N = 6089

Hit rate % = .71

a = 189.68

A = 941 Area % = .20

PAI = 3.55

6 Mo Robb KD

Map

n = 1132

N = 1663

Hit rate % = .68

a = 120.96 A = 941

Area % = .13

PAI = 5.23

6 Mo VT KD Map

n = 1329

N = 2200

Hit rate % = .6

a = 155.25 A = 941

Area % = .16

PAI = 3.75

9 Mo Asslt KD

Map

n = 2344

N = 3155

Hit rate % = .74

a = 167.19

A = 941 Area % = .18

PAI = 4.11

9 Mo Burg KD Map

n = 4381 N = 6089

Hit rate % = .72

a = 189.51

A = 941

Area % = .20

PAI = 3.6

9 Mo Robb KD

Map

n = 1215

N = 1663

Hit rate % = .73

a = 136.02

A = 941 Area % = .14

PAI = 5.21

9 Mo VT KD Map

n = 1447

N = 2200

Hit rate % = .66

a = 170.85

A = 941 Area % = .18

PAI = 3.67

12 Mo Asslt KD Map

n = 2384

N = 3155

Hit rate % = .76

a = 174.31

A = 941 Area % = .19

PAI = 4

12 Mo Burg KD Map

n = 4363

N = 6089

Hit rate % = .72

a = 189.4

A = 941 Area % = .20

PAI = 3.6

12 Mo Robb KD Map

n = 1229

N = 1663

Hit rate % = .74

a = 142.89

A = 941 Area % = .15

PAI = 4.93

12 Mo VT KD Map

n = 1497

N = 2200

Hit rate % = .68

a = 179.94

A = 941 Area % = .19

PAI = 3.58

9 Mo MCD (3/1/2010 -

11/30/2010)

9 Mo Assault

MCD

9 Mo Burglary

MCD

9 Mo Robbery

MCD

9 Mo Vehicle Theft

MCD

47

3 Mo Asslt KD

Map

n = 2655

N = 4638 Hit rate % = .57

a = 118.4 A = 941

Area % = .13

PAI = 4.38

3 Mo Burg KD Map

n = 6088

N = 9597

Hit rate % = .63

a = 162.43

A = 941 Area % = .17

PAI = 3.71

3 Mo Robb KD

Map

n = 1409

N = 2609 Hit rate % = .54

a = 92.61 A = 941

Area % = .1

PAI = 5.4

3 Mo VT KD Map

n = 1704

N = 3418 Hit rate % = .50

a = 124.33 A = 941

Area % = .13

PAI = 3.85

6 Mo Asslt KD

Map

n = 3186

N = 4638

Hit rate % = .69

a = 150.77 A = 941

Area % = .16

PAI = 4.31

6 Mo Burg KD Map

n = 6741

N = 9597

Hit rate % = .70

a = 189.68

A = 941 Area % = .20

PAI = 3.5

6 Mo Robb KD

Map

n = 1757

N = 2609

Hit rate % = .67

a = 120.96 A = 941

Area % = .13

PAI = 5.15

6 Mo VT KD Map

n = 2092

N = 3418

Hit rate % = .61

a = 155.25 A = 941

Area % = .16

PAI = 3.81

9 Mo Asslt KD

Map

n = 3405

N = 4638

Hit rate % = .73

a = 167.19

A = 941 Area % = .18

PAI = 4.06

9 Mo Burg KD Map

n = 6829 N = 9597

Hit rate % = .71

a = 189.51

A = 941

Area % = .20

PAI = 3.55

9 Mo Robb KD

Map

n = 1884

N = 2609

Hit rate % = .72

a = 136.02

A = 941 Area % = .14

PAI = 5.14

9 Mo VT KD Map

n = 2267

N = 3418

Hit rate % = .66

a = 170.85

A = 941 Area % = .18

PAI = 3.67

12 Mo Asslt KD Map

n = 3461

N = 4638

Hit rate % = .75

a = 174.31

A = 941 Area % = .19

PAI = 3.95

12 Mo Burg KD Map

n = 6807

N = 9597

Hit rate % = .71

a = 189.4

A = 941 Area % = .20

PAI = 3.55

12 Mo Robb KD Map

n = 1923

N = 2609

Hit rate % = .74

a = 142.89

A = 941 Area % = .15

PAI = 4.93

12 Mo VT KD Map

n = 2338

N = 3418

Hit rate % = .68

a = 179.94

A = 941 Area % = .19

PAI = 3.58

12 Mo MCD (3/1/2010 -

2/28/2011)

12 Mo Assault

MCD

12 Mo Burglary

MCD

12 Mo Robbery

MCD

12 Mo Vehicle Theft

MCD

48

3 Mo Asslt KD

Map

n = 3282

N = 5699 Hit rate % = .58

a = 118.4 A = 941

Area % = .13

PAI = 4.46

3 Mo Burg KD Map

n = 7691

N = 12178

Hit rate % = .63

a = 162.43

A = 941 Area % = .17

PAI = 3.71

3 Mo Robb KD

Map

n = 1793

N = 3349 Hit rate % = .54

a = 92.61 A = 941

Area % = .1

PAI = 5.4

3 Mo VT KD Map

n = 2272

N = 4533 Hit rate % = .50

a = 124.33 A = 941

Area % = .13

PAI = 3.85

6 Mo Asslt KD

Map

n = 3917

N = 5699

Hit rate % = .69

a = 150.77 A = 941

Area % = .16

PAI = 4.31

6 Mo Burg KD Map

n = 8560

N = 12178

Hit rate % = .70

a = 189.68

A = 941 Area % = .20

PAI = 3.5

6 Mo Robb KD

Map

n = 2228

N = 3349

Hit rate % = .67

a = 120.96 A = 941

Area % = .13

PAI = 5.15

6 Mo VT KD Map

n = 2792

N = 4533

Hit rate % = .62

a = 155.25 A = 941

Area % = .16

PAI = 3.88

9 Mo Asslt KD

Map

n = 4179

N = 5699

Hit rate % = .73

a = 167.19

A = 941 Area % = .18

PAI = 4.06

9 Mo Burg KD Map

n = 8667 N = 12178

Hit rate % = .71

a = 189.51

A = 941

Area % = .20

PAI = 3.55

9 Mo Robb KD

Map

n = 2390

N = 3349

Hit rate % = .71

a = 136.02

A = 941 Area % = .14

PAI = 5.07

9 Mo VT KD Map

n = 3009

N = 4533

Hit rate % = .66

a = 170.85

A = 941 Area % = .18

PAI = 3.67

12 Mo Asslt KD Map

n = 4257

N = 5699

Hit rate % = .75

a = 174.31

A = 941 Area % = .19

PAI = 3.95

12 Mo Burg KD Map

n = 8647

N = 12178

Hit rate % = .71

a = 189.4

A = 941 Area % = .20

PAI = 3.55

12 Mo Robb KD Map

n = 2471

N = 3349

Hit rate % = .74

a = 142.89

A = 941 Area % = .15

PAI = 4.93

12 Mo VT KD Map

n = 3115

N = 4533

Hit rate % = .69

a = 179.94

A = 941 Area % = .19

PAI = 3.63

49

Appendix C

Mean PAI Values for each Hotspot Mapping Technique


3 Mo Assault 3 Mo Burglary 3 Mo Robbery 3 Mo Vehicle Theft PAI Totals PAI Averages

PAI Average for

Hotspot

Mapping

Technique

2 2 4 2 10

3 3 4.5 2 12.5

5.5 2.4 6 2.33 16.23

4.66 3 5.33 2.25 15.24

53.97 53.97 / 16 = 3.37

6 Mo Assault 6 Mo Burglary 6 Mo Robbery 6 Mo Vehicle Theft

2 2 4 3 11

3.5 3 4.5 2.5 13.5

5 2.6 6.5 2.33 16.43

4.33 3 5.33 2.25 14.91

55.84 55.84 / 16 = 3.49


3 2 4 3 12

3.5 3 4.5 2.5 13.5 14.07 / 4 = 3.52

5.5 2.6 6.5 2.67 17.27

4.66 3 5.33 2.25 15.24

58.01 58.01 / 16 = 3.63


3 2 4 3 12

3.5 3 4.5 2.5 13.5

50

5.5 2.6 6.5 2.33 16.93

4.33 3 5.33 2.25 14.91

57.34 57.34 / 16 = 3.58


3 Mo Assault MCD 3 Mo Burglary MCD 3 Mo Robbery MCD 3 Mo Vehicle Theft MCD PAI Totals PAI Averages

PAI Average for

Hotspot

Mapping

Technique

30 10 10 6 56

40 20 30 10 100

25 40 60 20 145

40 25 80 30 175

476 476 / 16 = 29.75

6 Mo Assault MCD 6 Mo Burglary MCD 6 Mo Robbery MCD 6 Mo Vehicle Theft MCD 30 10 10 4

54 30 20 30 10

90 25 40 50 20

135 40 25 80 20

165

444 444 / 16 = 27.75


54 40 20 40 10

110 115.5 / 4 = 28.88

25 40 50 20 135

40 25 80 20 165

464 464 / 16 = 29


54 40 20 40 10

110

51

25 40 50 20 135

40 25 80 20 165

464 464 / 16 = 29

Hotspot Mapping Technique: Kernel Density (KD)

3 Mo Assault MCD 3 Mo Burglary MCD 3 Mo Robbery MCD 3 Mo Vehicle Theft MCD PAI Totals PAI Averages

PAI Average for

Hotspot

Mapping

Technique

4.46 3.76 5.6 4 17.82

4.38 3.55 5.23 3.88 17.04

4.17 3.6 5.21 3.72 16.7

4 3.6 4.93 3.63 16.16

67.72 67.72 / 16 = 4.23

6 Mo Assault MCD 6 Mo Burglary MCD 6 Mo Robbery MCD 6 Mo Vehicle Theft MCD 4.38 3.76 5.5 3.85

17.49 4.38 3.55 5.23 3.75

16.91 4.11 3.6 5.21 3.67

16.59 4 3.6 4.93 3.58

16.11

67.1 67.1 / 16 = 4.19


17.34 4.31 3.5 5.15 3.81

16.77 16.75 / 4 = 4.19

4.06 3.55 5.14 3.67 16.42

3.95 3.55 4.93 3.58 16.01

66.54 66.54 / 16 = 4.16


17.42 4.31 3.5 5.15 3.88

16.84

52

4.06 3.55 5.07 3.67 16.35

3.95 3.55 4.93 3.63 16.06

66.67 66.67 / 16 = 4.17

53

Appendix D

Mean PAI Values for each Crime Type


Measurement Crime Data (MCD) over Input

Crime Data (ICD) Assault Burglary Robbery Vehicle Theft

3 Mo MCD (3/1/2010 - 5/31/2010)

3 Mo ICD Hotspot Map 2 2 4 2

6 Mo ICD Hotspot Map 3 3 4.5 2

9 Mo ICD Hotspot Map 5.5 2.4 6 2.33

12 Mo ICD Hotspot Map 4.66 3 5.33 2.25

6 Mo MCD (3/1/2010 - 8/31/2010)



9 Mo ICD Hotspot Map 5 2.6 6.5 2.33


9 Mo MCD (3/1/2010 - 11/30/2010)



9 Mo ICD Hotspot Map 5.5 2.6 6.5 2.67


12 Mo MCD (3/1/2010 - 2/28/2011)








3 Mo MCD (3/1/2010 - 5/31/2010)





6 Mo MCD (3/1/2010 - 8/31/2010)





9 Mo MCD (3/1/2010 - 11/30/2010)





12 Mo MCD (3/1/2010 - 2/28/2011)

54





Hotspot Mapping Technique: Kernel Density

(KD)



3 Mo MCD (3/1/2010 - 5/31/2010)

3 Mo ICD Hotspot Map 4.46 3.76 5.6 4




6 Mo MCD (3/1/2010 - 8/31/2010)





9 Mo MCD (3/1/2010 - 11/30/2010)





12 Mo MCD (3/1/2010 - 2/28/2011)





PAI Totals 660.34 479.84 873.83 327.18

PAI Average for Hotspot Mapping Technique

660.34 /

48 =

13.76

479.84 / 48

= 10

873.83 / 48 =

18.2

327.18 / 48 =

6.82

55

Bibliography

Buerger, M. E., Cohn, E. G., & Petrosino, A. J. (1995). Defining the “hot spots of crime”:

Operationalizing theoretical concepts for field research. Crime and place, 4, 237-

257.

Bureau, U. S. C. (2014). State and County QuickFacts. Retrieved from

http://quickfacts.census.gov/qfd/states/18/1836003.html.

Chainey, S., & Ratcliffe, J. (2005). GIS and crime mapping. Chichester, West Sussex,

England ; Hoboken, NJ: Wiley.

Chainey, S., Tompson, L., & Uhlig, S. (2008). The Utility of Hotspot Mapping for

Predicting Spatial Patterns of Crime. Security Journal, 21(1-2), 4-28. doi: DOI

10.1057/palgrave.sj.8350066

Chamard, S. (2006). The history of crime mapping and its use by american police

departments. Alaska Justice Forum, 23, 1-8.

Eck, J. E., & Justice, N. I. o. (2005). Mapping crime understanding hot spots NIJ special

report (pp. v, 72 p.). Retrieved from http://purl.access.gpo.gov/GPO/LPS66131

Eck, J. E., & National Institute of Justice (U.S.). (2005). Mapping crime understanding

hot spots NIJ special report (pp. v, 72 p.). Retrieved from

http://purl.access.gpo.gov/GPO/LPS66131

FBI. (2004). Uniform Crime Reporting Handbook. Retrieved from

http://www2.fbi.gov/ucr/handbook/ucrhandbook04.pdf.

Harries, K. D., & National Institute of Justice (U.S.). (1999). Mapping crime : principle

and practice. Washington, DC: U.S. Dept. of Justice, Office of Justice Programs,

National Institute of Justice.

Hart, T., & Zandbergen, P. (2014). Kernel density estimation and hotspot mapping.

Policing, 37(2), 305-323.

Hengl, T. (2006). Finding the right pixel size. Computers & Geosciences, 32(9), 1283-

1298. doi: http://dx.doi.org/10.1016/j.cageo.2005.11.008

Heywood, D. I., Cornelius, S., & Carver, S. (1998). An introduction to geographical

information systems. New York: Addison Wesley Longman.

Kennedy, L. W., Caplan, J. M., & Piza, E. (2011). Risk Clusters, Hotspots, and Spatial

Intelligence: Risk Terrain Modeling as an Algorithm for Police Resource

Allocation Strategies. Journal of Quantitative Criminology, 27(3), 339-362. doi:

http://dx.doi.org/10.1007/s10940-010-9126-2

MacEachren, A. M. (1995). How maps work : representation, visualization, and design.

New York: Guilford Press.

Perry, W. L., McInnis, B., Price, C. C., Smith, S. C., & Hollywood, J. S. (2013).

Predictive policing : the role of crime forecasting in law enforcement operations.

PredPol. (2014). PredPol. Retrieved 03/01/2015, from http://www.predpol.com/results/

Reaves, B. A. (2010). Local police departments, 2007.: Retrieved from

http://www.bjs.gov/content/pub/pdf/lpd07.pdf.

Santos, R. B. (2005). Crime analysis and crime mapping. Thousand Oaks, Calif.: Sage

Publications.

Sherman, L. W., Gartin, P. R., & Buerger, M. E. (1989). Hot spots of predatory crime:

Routine activities and the criminology of place. Criminology, 27(1), 27.

http://quickfacts.census.gov/qfd/states/18/1836003.html



http://www2.fbi.gov/ucr/handbook/ucrhandbook04.pdf

http://dx.doi.org/10.1016/j.cageo.2005.11.008

http://dx.doi.org/10.1007/s10940-010-9126-2

http://www.predpol.com/results/

http://www.bjs.gov/content/pub/pdf/lpd07.pdf

56

Silverman, B. W. (1986). Density estimation for statistics and data analysis. London ;

New York: Chapman and Hall.

Weir, R., & Bangs, M. (2007). The use of Geographic Information Systems by crime

analysts in England and Wales: Citeseer.

Weisburd, D., & McEwen, T. (1998). Crime mapping and crime prevention. Monsey,

N.Y.: Criminal Justice Press.

Wickham, H. (2014). Tidy data. Journal of Statistical Software, 59(10), 1-23.

CURRICULUM VITAE


Education

Indiana University-Purdue University Indianapolis May 2011-May 2015

Master of Science in Geographic Information Science

Boston University January 2006-March 2007

Financial Planning Online Certificate Program

Colorado Christian University August 2001-May 2005

Bachelor of Arts in Global Studies

Nizhni Novgorod State University January-April 2004

Russian Study Abroad Program

Work Experience

Study Abroad Advisor October 2007-present

Indiana University-Purdue University Indianapolis

Advise a diverse student body on dozens of university sponsored study abroad

programs

Facilitate process of determining the transfer of credit for students studying

abroad through an outside institution

Coordinate with over 50 study abroad faculty directors on program promotion and

application processing

Monitor the processing of hundreds of study abroad applications per year

Send around 400 students abroad per year

Experience in managing international crises

Coordinate and conduct over 50 classroom presentations per year promoting study

abroad

Administer pre-departure orientation sessions to prepare students traveling abroad

Oversee the daily operations of the Study Abroad Office

Organized and facilitated study abroad promotional events attended by hundreds

of students per year

Assistant Language Teacher July 2005-July 2007

Yamagata, Japan Prefectural Board of Education

Taught English to over 800 Japanese students at two high schools

Responsible for preparing and administrating English lessons for 25 classes

Created and administered international awareness classes.

Memberships

NAFSA: Association of International Educators October 2007-present

The Forum on Education Abroad October 2009-present

Conferences

CIEE Annual Conference November 2013

Minneapolis, Minnesota

Forum on Education Abroad Annual Conference April 2013

Chicago, Illinois

Forum on Education Abroad: The Code of Ethics September 2012

Indianapolis, Indiana

Forum on Education Abroad Annual Conference March 2012

Denver, Colorado

NAFSA Region VI Conference November 2011

Louisville, Kentucky



NAFSA Indiana State Meeting June 2010

Bloomington, Indiana

Presented on “Hot Topics in Education Abroad”


Cincinnati, Ohio


Lexington, Kentucky

Presented on promoting study abroad using social media

NAFSA Annual Conference May 2008

Washington D.C.



Community Involvement

International Homestay Host


Hosted Swiss ESL student September 2010-February 2011

Hosted Japanese university student August-December 2009

Coordinator November 2006 & November 2007

Asahi-machi, Japan

Co-creator and organizer of largest cross-cultural event in Asahi-machi, over 200

attendees

Delegated jobs and managed the work of 25 volunteers

Tripled international monetary support from prior year

Workshop Instructor August 2006 & November 2006

Yamagata, Japan

Yamagata Prefectural Mid-Year Seminar: Presented "Effective

Internationalization"

Yamagata Prefectural ALT Orientation Seminar: Presented “Making the Most of

Your Yen”

45% of seminar attendees chose this workshop over 5 other workshops

Workshop evaluated as “best and most helpful”

Adult Community English Teacher January-June 2006

Asahi-machi, Japan

Documents

PREDICTIVE POLICING: A COMPARATIVE STUDY OF THREE …